The systems engineering of a network-centric distributed intelligent system of systems for robust human behavior classifications.

摘要

Automating intelligence within sensor networks for situational awareness and responses is the overall motivational application for this dissertation. Traditionally, intelligence is manually gathered and extracted by intelligence analysts. However, there will never be enough intelligence analysts, intelligent centers, or even bandwidth (for mobile sensors) to manually extract information for intelligence from raw sensor data. Fusing a large number of sensor types and inputs is also required. All of this can be implemented and automated in an artificial intelligent (AI) hierarchy described herein, and therefore not require human power to observe, fuse, and interpret. This objective is fulfilled in this systems dissertation with several independent systems combined together to form an intelligent system of systems (SoS).;In order to design and implement an intelligent SoS, there are a number of unique contributions from this author in this dissertation. The first six listed author contributions are systems developments as Chief Engineer on the intelligent SoS and the last six contributions are novel technological developments. The following are the SoS systems developments: (1) a Fixed Camera System containing a multi-camera network (thirty-six PoE cameras) and six processing units; (2) a Kiosk System containing dual Pan/Tilt/Zoom cameras, a microphone network and two processing units; and (3) a Command and Control System containing a database on a server with dual monitors displaying an (4) interactive executive graphical user interface displaying (5) mustered personnel and (6) abnormal behavior alarms. This SoS was designed and built with novel technologies that the author developed for this SoS: (7) high-level syntactical classifiers for classifying human/object behaviors that are predefined based on sequences of (8) identified combinations of fused (9) object recognitions (e.g. body postures and face recognitions) by low-level classifiers on video data, including a (10) generalized parts-based object recognition low-level classifier. The system uses a (11) high-level syntactical classifier to recover from low-level classification errors. This intelligent SoS was built and implemented as a prototype. Additionally, preliminary transitions are underway for transitioning the prototype to a product system, such as (12) providing a Field Programmable Gate Array (FPGA) architecture for the generalized object recognition low-level classifier.